hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Owen O'Malley (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HADOOP-134) JobTracker trapped in a loop if it fails to localize a task
Date Fri, 14 Apr 2006 23:42:00 GMT
     [ http://issues.apache.org/jira/browse/HADOOP-134?page=all ]

Owen O'Malley updated HADOOP-134:
---------------------------------

    Attachment: task-startup-safety.patch

Ok, this patch fixes the problem. In particular,
  1. adds the hostname to the task tracker names.
  2. moves exception-raising code out of the constructor for TaskInProgress
  3. pulls the task start up code into a separate procedure so that I can make sure it doesn't
throw any exceptions
  4. records failures during initialization of tasks
  5. moves the stringifyException into the utils package.

> JobTracker trapped in a loop if it fails to localize a task
> -----------------------------------------------------------
>
>          Key: HADOOP-134
>          URL: http://issues.apache.org/jira/browse/HADOOP-134
>      Project: Hadoop
>         Type: Bug

>   Components: mapred
>     Versions: 0.1.0
>     Reporter: Runping Qi
>  Attachments: task-startup-safety.patch
>
> The symptoms:
>     When I ran  jobs on a big cluster, I noticed that some jobs got stucked. Some map
tasks never got started. When I look at the log of the task tracker responsible for the tasks,
I saw the following exceptions:
> 060413 160702 Lost connection to JobTracker [kry1040/72.30.116.100:50020].  Retrying...
> java.io.IOException: No valid local directories in property: mapred.local.dir
>         at org.apache.hadoop.conf.Configuration.getFile(Configuration.java:282)
>         at org.apache.hadoop.mapred.JobConf.getLocalFile(JobConf.java:127)
>         at org.apache.hadoop.mapred.TaskTracker$TaskInProgress.localizeTask(TaskTracker.java:391)
>         at org.apache.hadoop.mapred.TaskTracker$TaskInProgress.<init>(TaskTracker.java:383)
>         at org.apache.hadoop.mapred.TaskTracker.offerService(TaskTracker.java:270)
>         at org.apache.hadoop.mapred.TaskTracker.run(TaskTracker.java:336)
>         at org.apache.hadoop.mapred.TaskTracker.main(TaskTracker.java:756)
> The reason for the exception is that the directory hadoop/mapred/local has "wrong" owner,
thus the task tracker cannot access to it.
> This caused the task tracker stucked into the following loops:
>             while (running) {
>                 boolean staleState = false;
>                 try {
>                     // This while-loop attempts reconnects if we get network errors
>                     while (running && ! staleState) {
>                         try {
>                             if (offerService() == STALE_STATE) {
>                                 staleState = true;
>                             }
>                         } catch (Exception ex) {
>                             LOG.log(Level.INFO, "Lost connection to JobTracker [" + jobTrackAddr
+ "].  Retrying...", ex);
>                             try {
>                                 Thread.sleep(5000);
>                             } catch (InterruptedException ie) {
>                             }
>                         }
>                     }
>                 } finally {
>                     close();
>                 }
>                 LOG.info("Reinitializing local state");
>                 initialize();
>             }
> Issue 1:
>     Method offerService() must catch and handle the exceptions that may be thrown from
new TaskInProgress() call, and report back to the job tracker if it cannot run the task. This
way, the task can be assigned to other task tracker.
> Issue 2:
>     The taskTracker should check whether it can access to the local dir at the initialization
time, before taking any tasks.
> Runping

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
   http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
   http://www.atlassian.com/software/jira


Mime
View raw message